1. Import Libraries

2. Load Dataset and Exploratory Data Analysis

3. Basic Data Cleaning

4. Dataset Description

https://community.ibm.com/accelerators/?context=analytics&query=telco%20customer&type=Data

This dataset was sourced from IBM and contains customer information from a Telecommunications company (Telco) that provides a variety of phone and internet services in California. Each row represents a customer and the columns provide different attributes of each customer such as demographic information, the type of services they are enrolled in, and whether or not they churned (i.e. left the company).

**Variables of Consideration**

  1. Churn_Value: Binary - 1 if customer churns, 0 if not

  2. Senior_Citizen: Binary - 1 if senior citizen, 0 if not

  3. Tech_Support: Binary - 'Yes' if customer has tech support, 'No' if not

  4. Internet_Service: Categorical - Type of internet service customer has (Fiber Optic, DSL, No internet service)

  5. Contract: Categorical - Customer contract types (month-to-month, one year, two year)

  6. Churn_Reason: Categorical - Reason why customer churned (ex. Competitor had better devices)

  7. City: Categorical - City customer resides in

  8. Population: Continuous - Population of city customer resides in

  9. Latitude: Continuous - Latitude of city customer resides in

  10. Monthly Charges: Continuous - Total monthly charges per customer

5. Dataset Findings

Given Telco is observing higher churn rates than the industry benchmark (27% vs 22%), our overall goal is to make recommendations as to how the company can improve churn rates, which will help improve customer retention and mitigate revenue loss.

Finding 1: Contract_Month-to-month and Internet Service_Fiber optic are key drivers as to why customers churn.

How did we get to our finding?

Run ML Decision Tree to find patterns in data

The decision tree pointed to Contract_Month-to-Month, Internet Service_Fiber optic, and Monthly Charges. We explored those variables in tables and plots below to validate the tree's findings.

Mid-range Goal - Based off the results from our Decision Tree, the primary reason customers will churn is if they have month-to-month contracts and subscribed to the Fiber Optic internet service. While month-to-month customers tend to churn more often due to lack of contractual obligation, having such a large proportion of customers churn off Fiber Optic indicates an issue with this product whether it be pricing, performance, or competitiveness with other service providers. Our recommendation to management is to work with product teams to investigate whether Fiber Optic quality and performance is meeting industry standards. We seek to explore more specific issues with Fiber Optic in Finding #2 to make more specific recommendations.

Finding 2: Dividing the service area into 4 bins based on latitude allowed us to identify thematic churn reasons for each binned region. We learned that by cutting the service area into different territories, there were different, prevalent issues occuring in each territory.

How did we get to our finding?

We tried to find patterns of churn, by churn categories

Then used regression tree to look for areas with highest churn.

Regression indicates high churn rate occurs in the area that has population greater than 27,249 which leads us to explore areas with the most churn.

Find areas with highest churn count and map them; the dot size represents count of churned customers.

Drill into zip codes with highest churn rates, top 40

Using the maps we decided to bin the service area by latitude to find patterns by region.

Use "Churn Reason" to get more granular and find specific patterns

Analyze the table of churn reasons by binned geographical to make informed business recommendations.

Long-term Goal -

Bin (37, 42] Northern California (includes Bay Area): "Competitor had Better Devices" is the primary churn reason. Recommend investing in improving device quality, in addition to upselling customers to better performance devices as consumers in this area are less price sensitive.

Bin (35, 37] Upper Central California: "Competitor had Better Devices" and "Attitude of Support Person" are the primary reasons for most churn. Recommend retraining the customer service and field agents that are customer facing in this region (i.e. sensitivity training) as well as upselling customers to better performance devices.

Bin (33, 35] Lower Central California (includes LA): "Attitude of Support Person", "Competitor had Better Devices", and "Competitor had Better Offer" were the top churn reasons. Recommend to use the same strategy of re-training customer service and field agents in this region in addition to bundling services and contracts for better deals, and upselling higher performance devices for those that want them.

Bin (30, 33] Southern California (includes SD): "Competitor Made Better Offer" is the primary churn reason. Recommend improving pricing of our services in this territory such as by offering deep discounts on bundling services or larger discounts for longer contracts.

Additional details on analysis

  1. Bin (37, 42] Northernmost California includes Bay Area: "Competitor had Better Devices" was main churn reason, accounts for 19% of all churn in area.
  2. Bin (35, 37] Central California: The combined churn reasons of "Attitude of Support Person" and "Attitude of Service Provider" account for 31% of the churn for the area.
  3. Bin (33, 35] Southern California incl. LA: This areas saw three major patterns in churn reason. The combined churn reasons of "Attitude of Support Person" and "Attitude of Service Provider" account for 21% of the churn, "Competitor had Better Devices" accounted for 16% of the churn and "Competitor Made Better Offer" made up 12% of the churn for the area making this region more complex than the rest.
  4. Bin (30, 33] Southernmost California incl. SD: "Competitor Made Better Offer" accounts for 69% of churn in the area. It is important to note that SD has predeominently Fiber Optic customers (74%).

Finding 3: Senior Citizens observe a higher churn rate, however this demographic observes less churn when they are subscribed to tech support. In addition, senior citizens with tech support have a higher average satisfaction score than non-senior citizens who have tech support indicating tech support has higher impact on the senior citizen population.

How did we get to our finding?

We looked for patterns based off of different demographics to check whether there were substantial differences between two populations and found senior citizens were noticeably different in their churn patterns than non-senior citizens. Our original hypothesis was that senior citizens were less likely to churn given the barriers and logistical efforts required to change service providers, however, we found this was not the case below.

Senior citizen population is more likely to churn than non-senior citizens (i.e. 41.68% vs 23.60%).

However, Senior citizens who have tech support are less likely to churn versus those who do not (i.e. 18.30% vs 4.46%).

Non Senior Citizens who don't have tech support observe half the churn rate of Senior Citizens who don't have tech support (i.e. 19.217% vs 37.215%)

Senior citizens who have tech support have a higher average satisfaction rate than not (3.303 vs 2.815).

Non-senior citizens who don't have tech support have a higher average satisfaction rate than senior citizens who don't (3.29 vs 2.815).

Senior citizens also have a higher concentration of fiber optic subscriptions than non-senior citizens (82.56% vs 64.89%). As we have previously mentioned, the fiber optic product is one of the primary reasons for churn.

Short-term Goal - Given this nuance in our senior citizen population, we recommend creating a new internet plan tailored towards seniors by automatically including free tech support along with their chosen internet service or bundling a tech support add-on at a significant discount. Senior citizens who stayed make up 17% of total current customers monthly revenue and senior citizens who churned make up 28% of the total churned customers monthly forgoed revenue hence improving retention in this demographic is important to overall company financial performance.

</div>